I have been using a simple low cost, high performance alternate solution to the standard Dual Mixer/DMTD. The idea is based on an analog version of NIST's "Tight Phase-Lock Loop Method of measuring Freq stability". http://tf.nist.gov/phase/Properties/one.htm#oneone fig 1.7 By replacing the "Voltage to freq converter, Freq counter & Printer with a PC data logging DVM, It was simple enough to be up and running from scratch in minutes, and the best part, it cost me nothing because I already had the four main parts that are needed. When a high resolution data logging DVM is used you don't need the offset voltage. To get better performance which seems to exceed most DMTD for low tau numbers, it takes a little more work and the use of a higher speed oversampling ADC data logger and a good offset voltage. I'll also add that this is not a popular solution, but as far as I've been able to determine it is the BEST SIMPLE configuration, IF you know analog and have an HP 10811 osc to use for the reference. I've attached a Basic modified NIST Block Diagram showing what I made: The NIST paper sums it up quite nicely: Using this configuration, it is not difficult to achieve a sensitivity of a part in 1e-14 per Hz resolution of the frequency counter, so one has excellent precision capabilities with this system. (I'm achieving well under 0.1 ps Phase resolution, and 1e-12 at Freq resolution with 0.1 TC, limited by my noisy reference) Note that the logged data is in Frequency and not Phase. I have found Ulrich's Plotter program great for doing the ADEV graphs As always, Negative criticism welcome, Have fun ws

WarrenS wrote: > I have been using a simple low cost, high performance alternate solution to the standard Dual Mixer/DMTD. > The idea is based on an analog version of NIST's "Tight Phase-Lock Loop Method of measuring Freq stability". > http://tf.nist.gov/phase/Properties/one.htm#oneone fig 1.7 > > By replacing the "Voltage to freq converter, Freq counter & Printer with a PC data logging DVM, > It was simple enough to be up and running from scratch in minutes, > and the best part, it cost me nothing because I already had the four main parts that are needed. > When a high resolution data logging DVM is used you don't need the offset voltage. > To get better performance which seems to exceed most DMTD for low tau numbers, it takes a little more work > and the use of a higher speed oversampling ADC data logger and a good offset voltage. > > I'll also add that this is not a popular solution, > but as far as I've been able to determine it is the BEST SIMPLE configuration, > IF you know analog and have an HP 10811 osc to use for the reference. > I've attached a Basic modified NIST Block Diagram showing what I made: > > The NIST paper sums it up quite nicely: > Using this configuration, it is not difficult to achieve a sensitivity of a part in 1e-14 per Hz resolution of the frequency counter, > so one has excellent precision capabilities with this system. > (I'm achieving well under 0.1 ps Phase resolution, and 1e-12 at Freq resolution with 0.1 TC, limited by my noisy reference) > > Note that the logged data is in Frequency and not Phase. > I have found Ulrich's Plotter program great for doing the ADEV graphs > > As always, Negative criticism welcome, > Have fun > ws > > Warren You need to ensure that the isolation between the 2 sources is sufficient to ensure that locking due to unwanted signal injection doesn't significantly effect the effective VCO EFC to frequency transfer function. This method requires that the one of the 2 sources being compared can be phase locked to the other. This isn't always possible, for example, if one wishes to evaluate the stability of an offset oscillator this technique cant be used. Bruce

Bruce Yes, there are a few disadvantages using this simple low cost configuration along with all of its advantages. You can not get everything for nothing, but you can get higher speed, better resolution and less noise from this. > This can not be used to evaluate the stability of an offset oscillator. Correct, It can not test osc frequencies that are much different than the ref osc such as an offset Osc, BUT one of the many advantages of this configuration is that an offset Osc is not needed. This will give freq offset data down to under 1 ms (>1 kHz) without Any offset osc. >You need to ensure that the isolation between the 2 sources is sufficient I have no problem with signal injection at 1e-12 resolution levels, even with 10811s without buffers using 10db attenuators and ground Isolation transformers at each Osc output. Maybe due to the fact that they are locked. This is verified by adding a slow low level freq modulation on the DUT osc and plotting the freq change on the reference's EFC. The phase error can also be changed by adding large or small offsets at the Loop amp's input to check for any pulling tendency. ws *********** From: "Bruce Griffiths" <bruce.griffiths@xtra.co.nz> > Warren > > You need to ensure that the isolation between the 2 sources is > sufficient to ensure that locking due to unwanted signal injection > doesn't significantly effect the effective VCO EFC to frequency transfer > function. > > This method requires that the one of the 2 sources being compared can be > phase locked to the other. > This isn't always possible, for example, if one wishes to evaluate the > stability of an offset oscillator this technique cant be used. > > Bruce > ************* > WarrenS wrote: >> I have been using a simple low cost, high performance alternate solution >> to the standard Dual Mixer/DMTD. >> The idea is based on an analog version of NIST's "Tight Phase-Lock Loop >> Method of measuring Freq stability". >> http://tf.nist.gov/phase/Properties/one.htm#oneone fig 1.7 >> >> By replacing the "Voltage to freq converter, Freq counter & Printer with >> a PC data logging DVM, >> It was simple enough to be up and running from scratch in minutes, >> and the best part, it cost me nothing because I already had the four main >> parts that are needed. >> When a high resolution data logging DVM is used you don't need the offset >> voltage. >> To get better performance which seems to exceed most DMTD for low tau >> numbers, it takes a little more work >> and the use of a higher speed oversampling ADC data logger and a good >> offset voltage. >> >> I'll also add that this is not a popular solution, >> but as far as I've been able to determine it is the BEST SIMPLE >> configuration, >> IF you know analog and have an HP 10811 osc to use for the reference. >> I've attached a Basic modified NIST Block Diagram showing what I made: >> >> The NIST paper sums it up quite nicely: >> Using this configuration, it is not difficult to achieve a sensitivity of >> a part in 1e-14 per Hz resolution of the frequency counter, >> so one has excellent precision capabilities with this system. >> (I'm achieving well under 0.1 ps Phase resolution, and 1e-12 at Freq >> resolution with 0.1 TC, limited by my noisy reference) >> >> Note that the logged data is in Frequency and not Phase. >> I have found Ulrich's Plotter program great for doing the ADEV graphs >> >> As always, Negative criticism welcome, >> Have fun >> ws >>

WarrenS wrote: > > Bruce > > Yes, there are a few disadvantages using this simple low cost > configuration > along with all of its advantages. > You can not get everything for nothing, but you can get higher speed, > better > resolution and less noise from this. > >> This can not be used to evaluate the stability of an offset oscillator. > Correct, It can not test osc frequencies that are much different than > the > ref osc such as an offset Osc, > BUT one of the many advantages of this configuration is that an offset > Osc > is not needed. > This will give freq offset data down to under 1 ms (>1 kHz) without Any > offset osc. > You also need to measure the EFC slope at the operating point as the EFC transfer characteristic can be highly nonlinear. >> You need to ensure that the isolation between the 2 sources is >> sufficient > I have no problem with signal injection at 1e-12 resolution levels, even > with 10811s without buffers > using 10db attenuators and ground Isolation transformers at each Osc > output. > Maybe due to the fact that they are locked. That is the worst possible case for injection locking. > This is verified by adding a slow low level freq modulation on the DUT > osc > and plotting the freq change on the reference's EFC. Probably not a good test for injection locking as a small shift in frequency from equality rapidly attenuates the effective injection locking signal. Better test is to insert a very high reverse isolation amplifier between each ocxo and the mixer and see if that makes any difference. Also need to ensure that injection locking doesn't occur through injection via the EFC input. > The phase error can also be changed by adding large or small offsets > at the Loop amp's input to check for any pulling tendency. > > ws > *********** Bruce

ws Reply to Bruce > You also need to measure the EFC slope at the operating point as the EFC > transfer characteristic can be highly nonlinear. Yes there is lots of things that can be done wrong but Another one of this configuration's many advantages is that the operating range of both the EFC and the Phase detector is very small, typical under a millivolt, so nonlinearly is NOT a problem. To calibrate end to end so that everything is included, The DUT can be changed by a small known offset, of say 1e-8 and measure the voltage change at the DVM/ADC output. Mine is calibrated for 1 mV per 1e-10 at the EFC, That calibration is linear over > than a 1 Hz (1e-7) offset range. > Also need to ensure that injection locking doesn't occur > through injection via the EFC input. Yes, One of the reasons for the isolations transformers (and lots of bypass caps). If there is ANY ground noise between the Oscillators it can effect the EFC voltage. Like all low level uV signal measurement and control, a lot of attention HAS to be paid to insure there is no added noise or errors. This takes good analog and digital understanding of possible noise sources. Differential input and output amps go a long way to insure there is no ground loops, offset voltages or noise coupled errors. >>Maybe due to the fact that the Osc are locked. > That is the worst possible case for injection locking. Possible, but not exactly what I've seen in this configuration. The effect of coupling between Oscillators is very phase sensitive, and can be positive, neg or null as their relative phase shifts. By adjusting the phase in the way I noted any changing effects can be easily seen when there is ANY interaction between Osc. >> This is verified by adding a slow low level freq modulation on the DUT > Probably not a good test for injection locking as a small shift in > frequency from equality rapidly attenuates the effective injection > locking signal. Another one of the advantages of this circuit is that there is NO shift in freq from equaqlity (If that means what I think it is). In any case, I have found this to be a good test because ANY coupling of any type between Oscillators causes a nonlinear transfer function, as a function of voltage and/or freq. So by checking that the transfer function between the DUT EFC input and the fast ADC output is linear and freq independent over a wide range of signals, non coupling is assured down to the level of the noise. Another check I did was to unlock the two Osc and add an freq offset, to see if that caused ANY effect at all on the other Osc. > Better test is to insert a very high reverse isolation amplifier between > each ocxo and the mixer and see if that makes any difference. Maybe so, and I did try to do that per your early suggestion, but the test was unsuccessful because of my poor isolation/buffer amps. It's something I'll get back to when I've lower the noise more to see if there is anything below the present noise level. ws *********************** From: "Bruce Griffiths" <bruce.griffiths@xtra.co.nz> > WarrenS wrote: >> >> Bruce >> >> Yes, there are a few disadvantages using this simple low >> cost configuration along with all of its advantages. >> You can not get everything for nothing, but you can get >> higher speed, better resolution and less noise from this. >> >>> This can not be used to evaluate the stability of an offset oscillator. >> Correct, It can not test osc frequencies that are much different than >> the ref osc such as an offset Osc, BUT >> one of the many advantages of this configuration is that an offset Osc >> is not needed. This will give freq offset data down to >> under 1 ms (>1 kHz) without Any offset osc. >> > You also need to measure the EFC slope at the operating point as the EFC > transfer characteristic can be highly nonlinear. > >>> You need to ensure that the isolation between the 2 sources is sufficient >> I have no problem with signal injection at 1e-12 resolution levels, even >> with 10811s without buffers using 10db attenuators and ground Isolation >> transformers at each Osc output. Maybe due to the fact that they are locked. > > That is the worst possible case for injection locking. > >> This is verified by adding a slow low level freq modulation on the DUT >> osc and plotting the freq change on the reference's EFC. > > Probably not a good test for injection locking as a small shift in > frequency from equality rapidly attenuates the effective injection > locking signal. > Better test is to insert a very high reverse isolation amplifier between > each ocxo and the mixer and see if that makes any difference. > Also need to ensure that injection locking doesn't occur through > injection via the EFC input. > > >> The phase error can also be changed by adding large or small offsets >> at the Loop amp's input to check for any pulling tendency. >> >> ws >> > Bruce > ******************** > From: "Bruce Griffiths" <bruce.griffiths@xtra.co.nz> >> Warren >> >> You need to ensure that the isolation between the 2 sources is >> sufficient to ensure that locking due to unwanted signal injection >> doesn't significantly effect the effective VCO EFC to frequency transfer >> function. >> >> This method requires that the one of the 2 sources being compared can be >> phase locked to the other. >> This isn't always possible, for example, if one wishes to evaluate the >> stability of an offset oscillator this technique cant be used. >> >> Bruce >> > ************* >> WarrenS wrote: >>> I have been using a simple low cost, high performance alternate solution >>> to the standard Dual Mixer/DMTD. >>> The idea is based on an analog version of NIST's "Tight Phase-Lock Loop >>> Method of measuring Freq stability". >>> http://tf.nist.gov/phase/Properties/one.htm#oneone fig 1.7 >>> >>> By replacing the "Voltage to freq converter, Freq counter & Printer with >>> a PC data logging DVM, >>> It was simple enough to be up and running from scratch in minutes, >>> and the best part, it cost me nothing because I already had the four main >>> parts that are needed. >>> When a high resolution data logging DVM is used you don't need the offset >>> voltage. >>> To get better performance which seems to exceed most DMTD for low tau >>> numbers, it takes a little more work >>> and the use of a higher speed oversampling ADC data logger and a good >>> offset voltage. >>> >>> I'll also add that this is not a popular solution, >>> but as far as I've been able to determine it is the BEST SIMPLE >>> configuration, >>> IF you know analog and have an HP 10811 osc to use for the reference. >>> I've attached a Basic modified NIST Block Diagram showing what I made: >>> >>> The NIST paper sums it up quite nicely: >>> Using this configuration, it is not difficult to achieve a sensitivity of >>> a part in 1e-14 per Hz resolution of the frequency counter, >>> so one has excellent precision capabilities with this system. >>> (I'm achieving well under 0.1 ps Phase resolution, and 1e-12 at Freq >>> resolution with 0.1 TC, limited by my noisy reference) >>> >>> Note that the logged data is in Frequency and not Phase. >>> I have found Ulrich's Plotter program great for doing the ADEV graphs >>> >>> As always, Negative criticism welcome, >>> Have fun >>> ws >>>

WarrenS wrote: > ws Reply to Bruce > > >> You also need to measure the EFC slope at the operating point as the EFC >> transfer characteristic can be highly nonlinear. >> > Yes there is lots of things that can be done wrong but > Another one of this configuration's many advantages is that the operating > range of both the EFC and the Phase detector is very small, > typical under a millivolt, so nonlinearly is NOT a problem. > To calibrate end to end so that everything is included, The DUT can be changed by a > small known offset, of say 1e-8 and measure the voltage change at the DVM/ADC output. > Mine is calibrated for 1 mV per 1e-10 at the EFC, > That calibration is linear over > than a 1 Hz (1e-7) offset range. > > > Whilst that may be true for your OCXO, this is certainly not true for every ocxo one may wish to measure. Since mixer nonlinearity near zero phase isnt an issue with a quiet ocxo pair, a traditional diode double balanced mixer phase detector would be appropriate as it hs lower noise than any other mixer/phase detector. >> Also need to ensure that injection locking doesn't occur >> through injection via the EFC input. >> > Yes, One of the reasons for the isolations transformers (and lots of bypass caps). > The isolation tranformers are not useful for blocking RF, nor in general is an arbitrary bypass capacitor array. > If there is ANY ground noise between the Oscillators it can effect the EFC voltage. > Like all low level uV signal measurement and control, a lot of attention HAS to be paid > to insure there is no added noise or errors. This takes good analog and digital understanding > of possible noise sources. Differential input and output amps go a long way to insure > there is no ground loops, offset voltages or noise coupled errors. > > These measures have little or no effect on RF injection via the EFC input. Effective filtering of the phase detector output and use of a high isolation mixer low noise is required. >>> Maybe due to the fact that the Osc are locked. >>> >> That is the worst possible case for injection locking. >> > Possible, but not exactly what I've seen in this configuration. > The effect of coupling between Oscillators is very phase sensitive, > and can be positive, neg or null as their relative phase shifts. > By adjusting the phase in the way I noted any changing effects > can be easily seen when there is ANY interaction between Osc. > > > I'd still prefer to compare results with and without the isolation amplifiers. >>> This is verified by adding a slow low level freq modulation on the DUT >>> >> Probably not a good test for injection locking as a small shift in >> frequency from equality rapidly attenuates the effective injection >> locking signal. >> > Another one of the advantages of this circuit is that there is NO shift in freq > from equaqlity (If that means what I think it is). In any case, > I have found this to be a good test because ANY coupling of any type between > Oscillators causes a nonlinear transfer function, as a function of voltage and/or freq. > So by checking that the transfer function between the DUT EFC input > and the fast ADC output is linear and freq independent over a wide range of signals, > non coupling is assured down to the level of the noise. > Another check I did was to unlock the two Osc and add an freq offset, > to see if that caused ANY effect at all on the other Osc. > > If there is another contributor to phase locking then the full frequency instability wont be reflected by the EFC input fluctuations. >> Better test is to insert a very high reverse isolation amplifier between >> each ocxo and the mixer and see if that makes any difference. >> > Maybe so, and I did try to do that per your early suggestion, but the test was unsuccessful > because of my poor isolation/buffer amps. It's something I'll get back to when > I've lower the noise more to see if there is anything below the present noise level. > > ws > > *********************** > From: "Bruce Griffiths" <bruce.griffiths@xtra.co.nz> > > > Bruce

ws reply to ">" Bruce's comments >> That calibration is linear over > than a 1 Hz (1e-7) offset range. > Whilst that may be true for your OCXO, this is certainly > not true for every ocxo one may wish to measure. It is not the measured OSC that needs to be linear, but the reference Osc, but I'm sure you knew that. And NO not all reference Osc will be that good, but close enough, probable by an order of magnitude when proper end to end calibration is correctly done. If the range is small enough, it will be linear enough even for a Very nonlinear EFC curve. This is built to check High precision 10 MHz Osc, If they are off by more a Hz or even 1e-9, you don't need this kind of performance. This will easy give 1% absolute difference accuracy at 10 MHz. So the freq difference for a 1e-10 reading could be from 0.99e-10 to 1.01e-10. (plus noise) Most would consider it great to be able to take ADEV numbers that are not off more than two to one. > Since mixer nonlinearity near zero phase isn't an issue with a quiet ocxo > pair, a traditional diode double balanced mixer phase detector would be > appropriate as it has lower noise than any other mixer/phase detector. With the high negative feedback that this has (2K+), the mixer hardly leaves 'zero' even with a noisy pair. I'm using a standard Mini-circuit DB Mixer. The mixer is working good enough so as not to be providing any noticeable error or noise. I've tried two other ones and they all give about the same results. The Mixer is inside the Loop with a lot of negative feedback around it, Much of its basic characteristics and error are reduced by the loop gain, including much of its noise. The system noise is mostly determined by Just the EFC input noise and little else. The effective isolation between Oscillators with the mixer output RF shorted and the 10dB attenuators, is good enough not to cause any Osc interaction. > The isolation transformers are not useful for blocking RF, Right, they pass RF, they don't block it. BUT The Isolation transformers are VERY important for a couple of reasons. The way these double oven 10811 are made, Their RF output is not isolated, but grounded at the inner case. At the end of the well shielded but long RF terminated cable there is a lot of RF noise on the cable shield ground AND a few mV of DC. If this Shield ground is Not isolated but connected to anything, It can cause a lot of errors. These errors can be orders of magnitude worse than other error sources. The RF shield ground on this OXCO can not be used in a dual Osc system, to get anywhere near max possible performance. It must be isolated. > nor in general is an arbitrary bypass capacitor array. The circuit needs bypass caps because of the low level, wide bandwidth, high gain, "DC", feedback loop. Even when isolated there is enough RF on the ground shield that it still acts pretty acts like a 40 db down transmitting antenna. That RF needs to be kept out of the uV wide band signals. (And the other Osc) > I'd still prefer to compare results with and without the isolation amplifiers. Yes, and I agree for a good safe universal GP design, They need to be there. I just don't happen to have any yet, and they are low on my priority list of needed improvements. With my specific setup, I have tested it well enough that I'm sure that their effect at this time would be small enough as not to be noticed. > If there is another contributor to phase locking then the full frequency > instability wont be reflected by the EFC input fluctuations. Don't get me wrong, in the beginning the oscillators pretty much acted like they were sync locked because of poor ground, common PS, non isolated RF output, etc,etc. That did not go away by luck. What I'm saying is they do not now know that each other exist to the level that I can test to, which is about 5e-13. Something that some seem to of missed is that this is not the standard RF circuit configuration with the standard open loop errors. This is a closed loop Neg feedback "PID type" freq control system where the errors inside the loop are reduced by the loop gain. The Only significant error outside the loop is the Osc its self and Osc's internal EFC offset. The EFC range being used during a measurement period is typically less than 1/1000 of its range. ws **************** ----- Original Message ----- From: "Bruce Griffiths" <bruce.griffiths@xtra.co.nz> To: "WarrenS" <warrensjmail-one@yahoo.com>; "Discussion of precise time and frequency measurement" <time-nuts@febo.com> Sent: Friday, October 02, 2009 6:00 PM Subject: Re: [time-nuts] Low cost alternate to Dual Mixer/DMTD > WarrenS wrote: >> ws Reply to Bruce >> >>> You also need to measure the EFC slope at the operating point as the EFC >>> transfer characteristic can be highly nonlinear. >>> >> Yes there is lots of things that can be done wrong but >> Another one of this configuration's many advantages is that the operating >> range of both the EFC and the Phase detector is very small, >> typical under a millivolt, so nonlinearly is NOT a problem. >> To calibrate end to end so that everything is included, The DUT can be changed by a >> small known offset, of say 1e-8 and measure the voltage change at the DVM/ADC output. >> Mine is calibrated for 1 mV per 1e-10 at the EFC, >> That calibration is linear over > than a 1 Hz (1e-7) offset range. >> > Whilst that may be true for your OCXO, this is certainly not true for > every ocxo one may wish to measure. > Since mixer nonlinearity near zero phase isnt an issue with a quiet ocxo > pair, a traditional diode double balanced mixer phase detector would be > appropriate as it hs lower noise than any other mixer/phase detector. >> >>> Also need to ensure that injection locking doesn't occur >>> through injection via the EFC input. >>> >> Yes, One of the reasons for the isolations transformers (and lots of bypass caps). >> > The isolation tranformers are not useful for blocking RF, nor in general > is an arbitrary bypass capacitor array. >> If there is ANY ground noise between the Oscillators it can effect the EFC voltage. >> Like all low level uV signal measurement and control, a lot of attention HAS to be paid >> to insure there is no added noise or errors. This takes good analog and digital understanding >> of possible noise sources. Differential input and output amps go a long way to insure >> there is no ground loops, offset voltages or noise coupled errors. >> > These measures have little or no effect on RF injection via the EFC input. > Effective filtering of the phase detector output and use of a high > isolation mixer low noise is required. >>>> Maybe due to the fact that the Osc are locked. >>>> >>> That is the worst possible case for injection locking. >>> >> Possible, but not exactly what I've seen in this configuration. >> The effect of coupling between Oscillators is very phase sensitive, >> and can be positive, neg or null as their relative phase shifts. >> By adjusting the phase in the way I noted any changing effects >> can be easily seen when there is ANY interaction between Osc. >> >> > I'd still prefer to compare results with and without the isolation > amplifiers. >> >>>> This is verified by adding a slow low level freq modulation on the DUT >>>> >>> Probably not a good test for injection locking as a small shift in >>> frequency from equality rapidly attenuates the effective injection >>> locking signal. >>> >> Another one of the advantages of this circuit is that there is NO shift in freq >> from equaqlity (If that means what I think it is). In any case, >> I have found this to be a good test because ANY coupling of any type between >> Oscillators causes a nonlinear transfer function, as a function of voltage and/or freq. >> So by checking that the transfer function between the DUT EFC input >> and the fast ADC output is linear and freq independent over a wide range of signals, >> non coupling is assured down to the level of the noise. >> Another check I did was to unlock the two Osc and add an freq offset, >> to see if that caused ANY effect at all on the other Osc. >> >> > If there is another contributor to phase locking then the full frequency > instability wont be reflected by the EFC input fluctuations. >> >>> Better test is to insert a very high reverse isolation amplifier between >>> each ocxo and the mixer and see if that makes any difference. >>> >> Maybe so, and I did try to do that per your early suggestion, but the test was unsuccessful >> because of my poor isolation/buffer amps. It's something I'll get back to when >> I've lower the noise more to see if there is anything below the present noise level. >> >> ws >> >> *********************** >> From: "Bruce Griffiths" <bruce.griffiths@xtra.co.nz> >> >> >> > Bruce > > >

> > If there is another contributor to phase locking then the full frequency > > instability wont be reflected by the EFC input fluctuations. > Don't get me wrong, in the beginning the oscillators pretty much > acted like they were sync locked > because of poor ground, common PS, non isolated RF output, > etc,etc. That did not go away by luck. > What I'm saying is they do not now know that each other exist to > the level that I can test to, which is about 5e-13. It's been my experience with the HP 3048A, which is basically the same thing as what you have built, that at least 20-30 dB of reverse isolation is necessary when two OCXOs (all 10811s and most of the Wenzel units I have) drive the LO and RF ports of a single mixer. Make no mistake, this is not something you can blow off in the general case. It'll be good to see some comparative results once you have some additional isolation in place. Have you tried modulating your DUT oscillator with white noise and FFT'ing the correction signal to visualize the actual loop bandwidth? My guess is that your effective LBW is somewhat wider than your design target, thanks to the injection-locking bugbear. That said, yes, your setup sounds like a very reasonable approach to better-than-TIC grade timing measurement, especially with a digital back end. It can certainly win the price-performance contest over other architectures. It's great to see several different approaches actually being tried, as opposed to speculated upon! > Something that some seem to of missed is that this is not the > standard RF circuit configuration with the standard open loop errors. > This is a closed loop Neg feedback "PID type" freq control system > where the errors inside the loop are reduced by the loop gain. > The Only significant error outside the loop is the Osc its self > and Osc's internal EFC offset. > The EFC range being used during a measurement period is typically > less than 1/1000 of its range. As Bruce points out, that doesn't matter in the least; in fact, I'd submit that injection locking is the very reason why you are seeing such a small EFC correction range. Ordinarily two oscillators won't track that well in response to shared environmental conditions alone. Fortunately injection locking is the easiest (and cheapest) malady to fix! Another interesting challenge will be to find a way to determine your true residual floor. Because your measurement is frequency-based, you presumably can't just feed both DUT and reference ports from a splitter and watch the residual phase error over time, correct? -- john, KE5FX

WarrenS wrote: > ws reply to ">" Bruce's comments > > >>> That calibration is linear over > than a 1 Hz (1e-7) offset range. >>> >> Whilst that may be true for your OCXO, this is certainly >> not true for every ocxo one may wish to measure. >> > It is not the measured OSC that needs to be linear, > but the reference Osc, but I'm sure you knew that. > And NO not all reference Osc will be that good, > but close enough, probable by an order of magnitude > when proper end to end calibration is correctly done. > If the range is small enough, it will be linear enough even > for a Very nonlinear EFC curve. > This is built to check High precision 10 MHz Osc, > If they are off by more a Hz or even 1e-9, you don't need this kind of performance. > Not true, if the offset is accurately known then a stable oscillator that has a frequency offset > 1E-9 is just as useful for calibration purposes as one that has an offset of 1E-11 or less. > This will easy give 1% absolute difference accuracy at 10 MHz. > So the freq difference for a 1e-10 reading could be from 0.99e-10 to 1.01e-10. (plus noise) > Most would consider it great to be able to take ADEV numbers that are not off more than two to one. > > >> Since mixer nonlinearity near zero phase isn't an issue with a quiet ocxo >> pair, a traditional diode double balanced mixer phase detector would be >> appropriate as it has lower noise than any other mixer/phase detector. >> > With the high negative feedback that this has (2K+), the mixer hardly leaves 'zero' even with a noisy pair. > I'm using a standard Mini-circuit DB Mixer. > The mixer is working good enough so as not to be providing any noticeable error or noise. > Have you actually measured the mixer output noise when used as a phase detector? > I've tried two other ones and they all give about the same results. > One can either use them with both the RF and LO ports saturated and suppress AM noise or only saturate the LO port. The latter mode of operation tends to have a lower mixer phase shift tempco with a lower phase sensitivity than when both the LO and RF ports are saturated. Phase detector IF port termination also affects its gain and noise. Terminating the IF port in a capacitive load reduces the noise and increases the phase detector gain. > The Mixer is inside the Loop with a lot of negative feedback around it, > Much of its basic characteristics and error are reduced by the loop gain, > including much of its noise. The system noise is mostly determined by > Just the EFC input noise and little else. > The effective isolation between Oscillators with the mixer output RF shorted and the 10dB attenuators, > is good enough not to cause any Osc interaction. > > I doubt it. Try estimating the required isolation using Adler's equation. >> The isolation transformers are not useful for blocking RF, >> > Right, they pass RF, they don't block it. BUT > The Isolation transformers are VERY important for a couple of reasons. > The way these double oven 10811 are made, Their RF output is not isolated, > but grounded at the inner case. At the end of the well shielded but long RF terminated cable > there is a lot of RF noise on the cable shield ground AND a few mV of DC. > If this Shield ground is Not isolated but connected to anything, It can cause a lot of errors. > These errors can be orders of magnitude worse than other error sources. > The RF shield ground on this OXCO can not be used in a dual Osc system, > to get anywhere near max possible performance. It must be isolated. > > >> nor in general is an arbitrary bypass capacitor array. >> > The circuit needs bypass caps because of the low level, wide bandwidth, high gain, "DC", feedback loop. > Even when isolated there is enough RF on the ground shield that it still acts > pretty acts like a 40 db down transmitting antenna. > That RF needs to be kept out of the uV wide band signals. (And the other Osc) > > > >> I'd still prefer to compare results with and without the isolation amplifiers. >> > Yes, and I agree for a good safe universal GP design, They need to be there. > I just don't happen to have any yet, and they are low on my priority list of needed improvements. > With my specific setup, I have tested it well enough that I'm sure that their > effect at this time would be small enough as not to be noticed. > > >> If there is another contributor to phase locking then the full frequency >> instability wont be reflected by the EFC input fluctuations. >> > Don't get me wrong, in the beginning the oscillators pretty much acted like they were sync locked > because of poor ground, common PS, non isolated RF output, etc,etc. That did not go away by luck. > What I'm saying is they do not now know that each other exist to the level that I can test to, which is about 5e-13. > > > Something that some seem to of missed is that this is not the standard RF circuit configuration with the standard open loop errors. > This is a closed loop Neg feedback "PID type" freq control system where the errors inside the loop are reduced by the loop gain. > The Only significant error outside the loop is the Osc its self and Osc's internal EFC offset. > The EFC range being used during a measurement period is typically less than 1/1000 of its range. > > ws > > **************** > ----- Original Message ----- > From: "Bruce Griffiths" <bruce.griffiths@xtra.co.nz> > To: "WarrenS" <warrensjmail-one@yahoo.com>; "Discussion of precise time and frequency measurement" <time-nuts@febo.com> > Sent: Friday, October 02, 2009 6:00 PM > Subject: Re: [time-nuts] Low cost alternate to Dual Mixer/DMTD > > > >> WarrenS wrote: >> >>> ws Reply to Bruce >>> >>> >>>> You also need to measure the EFC slope at the operating point as the EFC >>>> transfer characteristic can be highly nonlinear. >>>> >>>> >>> Yes there is lots of things that can be done wrong but >>> Another one of this configuration's many advantages is that the operating >>> range of both the EFC and the Phase detector is very small, >>> typical under a millivolt, so nonlinearly is NOT a problem. >>> To calibrate end to end so that everything is included, The DUT can be changed by a >>> small known offset, of say 1e-8 and measure the voltage change at the DVM/ADC output. >>> Mine is calibrated for 1 mV per 1e-10 at the EFC, >>> That calibration is linear over > than a 1 Hz (1e-7) offset range. >>> >>> >> Whilst that may be true for your OCXO, this is certainly not true for >> every ocxo one may wish to measure. >> Since mixer nonlinearity near zero phase isnt an issue with a quiet ocxo >> pair, a traditional diode double balanced mixer phase detector would be >> appropriate as it hs lower noise than any other mixer/phase detector. >> >>>> Also need to ensure that injection locking doesn't occur >>>> through injection via the EFC input. >>>> >>>> >>> Yes, One of the reasons for the isolations transformers (and lots of bypass caps). >>> >>> >> The isolation tranformers are not useful for blocking RF, nor in general >> is an arbitrary bypass capacitor array. >> >>> If there is ANY ground noise between the Oscillators it can effect the EFC voltage. >>> Like all low level uV signal measurement and control, a lot of attention HAS to be paid >>> to insure there is no added noise or errors. This takes good analog and digital understanding >>> of possible noise sources. Differential input and output amps go a long way to insure >>> there is no ground loops, offset voltages or noise coupled errors. >>> >>> >> These measures have little or no effect on RF injection via the EFC input. >> Effective filtering of the phase detector output and use of a high >> isolation mixer low noise is required. >> >>>>> Maybe due to the fact that the Osc are locked. >>>>> >>>>> >>>> That is the worst possible case for injection locking. >>>> >>>> >>> Possible, but not exactly what I've seen in this configuration. >>> The effect of coupling between Oscillators is very phase sensitive, >>> and can be positive, neg or null as their relative phase shifts. >>> By adjusting the phase in the way I noted any changing effects >>> can be easily seen when there is ANY interaction between Osc. >>> >>> >>> >> I'd still prefer to compare results with and without the isolation >> amplifiers. >> >>>>> This is verified by adding a slow low level freq modulation on the DUT >>>>> >>>>> >>>> Probably not a good test for injection locking as a small shift in >>>> frequency from equality rapidly attenuates the effective injection >>>> locking signal. >>>> >>>> >>> Another one of the advantages of this circuit is that there is NO shift in freq >>> from equaqlity (If that means what I think it is). In any case, >>> I have found this to be a good test because ANY coupling of any type between >>> Oscillators causes a nonlinear transfer function, as a function of voltage and/or freq. >>> So by checking that the transfer function between the DUT EFC input >>> and the fast ADC output is linear and freq independent over a wide range of signals, >>> non coupling is assured down to the level of the noise. >>> Another check I did was to unlock the two Osc and add an freq offset, >>> to see if that caused ANY effect at all on the other Osc. >>> >>> >>> >> If there is another contributor to phase locking then the full frequency >> instability wont be reflected by the EFC input fluctuations. >> >>>> Better test is to insert a very high reverse isolation amplifier between >>>> each ocxo and the mixer and see if that makes any difference. >>>> >>>> >>> Maybe so, and I did try to do that per your early suggestion, but the test was unsuccessful >>> because of my poor isolation/buffer amps. It's something I'll get back to when >>> I've lower the noise more to see if there is anything below the present noise level. >>> >>> ws >>> >>> *********************** >>> From: "Bruce Griffiths" <bruce.griffiths@xtra.co.nz> >>> >>> >>> >>> >> Bruce >> >> >> >> > >

Sorry, an incomplete statement on my part. I should of at least said

I've done some noise test on it but not what you would consider a standard noise test..

One of the BIGGER advantages of this configuration is that many of the 

Again Something I'd like to try and get across is many standard things like 

I put a cap right on its output as the first stage of the LP passive RF filter.

I don't know about Adler's equation and a lot of other things that would be nice to know

" ws", ">Bruce", ">>ws" ">>>Bruce" ">>>>ws" ">>>>>Bruce" >> If they are off you don't need this kind of performance. > Not true, if the offset is accurately known then a stable oscillator > that has a frequency offset > 1E-9 is just as useful for calibration > purposes as one that has an offset of 1E-11 or less. Sorry, an incomplete statement on my part. I should of at least said If it is CHANGING by more than a Hz or even 1e-9, ... As long as it's nominal freq is consistent and with-in range of the Reference Osc, then end to end calibration can be checked at any desired freq to minimize calibration errors which I've done by using an accurate (and somewhat noise) standard high resolution digital freq synthesizer. > Have you actually measured the mixer output noise when used as a phase detector? I've done some noise test on it but not what you would consider a standard noise test.. I've checked its wide band Peak to peak noise both open and closed loop. Also did an 'audio' spectrum sweep of the phase detectors output to insure that there are no spurs being caused by AM, FM, PS, or ground loops etc. Being a control loop person, the most reassuring test that I have done is to check the noise of the Reference Osc with and without the Feedback connected. This is a standard control loop test to insure that the loop is behaving correctly and not adding a lot of its own noise to the system. > One can either use them with both the RF and LO ports saturated and > suppress AM noise or only saturate the LO port. One of the BIGGER advantages of this configuration is that many of the standard things one normally has to consider do not directly effect this system. Take AM noise, this causes small changes in the loop gain, which effect even smaller changes in the loop TC which effect even smaller changes in the measured signal noise. The loop TC is set way above the alising filter's bandwidth of the ADC which is set way above the Tau 0 time. > The latter mode of operation tends to have a lower mixer phase shift > tempco with a lower phase sensitivity than when both the LO and RF ports > are saturated. Again Something I'd like to try and get across is many standard things like phase Tempco etc have NO significant effect on the measured DUT noise with this configuration because they are inside a negative feedback Frequency loop. I have verified the limited effect of phase noise by injecting larges amounts of signal error at the phase detector output. A one Hz sign wave is down more than 60 dB, and anything slower like TC or DC is much less. Also standard things that can drive a standard high resolution Phase measurement crazy such as touching a cable or connecter have very little or no effect because their effect is inside the loop and most all their freq content is outside the measured noise spectrum. > Phase detector IF port termination also affects its gain and noise. > Terminating the IF port in a capacitive load reduces the noise and > increases the phase detector gain. I put a cap right on its output as the first stage of the LP passive RF filter. This being a neg feedback loop where the main effect of the phase detector is to effects the loop gain, what I've done is to insure that the loop gain is reassuring constant by checking the open loop response of the Phase detector. If your point is that my setup is not optimized, I totally agree. I know that it can be made much better, I have not done all that can be done to optimize the phase detector noise, But for now it is low enough and much better that my reference Osc, so further improvements are both mostly unnecessary and hard to verify. >> The effective isolation between Oscillators with the mixer output RF shorted (by cap) >> and the 10dB attenuators, is good enough not to cause (any measurable) Osc interaction > I doubt it, Try estimating the required isolation using Adler's equation. I don't know about Adler's equation and a lot of other things that would be nice to know But I do know how to test for cause and effect. One of the early test I did was to connect the two Oscillators outputs together thru a variable attenuator to see how much isolation I'd need and to see the effects when there is not enough isolation. That is when I found to my surprise the need to keep the RF ground shields isolated thru ground isolation transformers. I'll stand by my statement and can verify with test, that for the current configuration that I am using, The two OSC have NO interaction which is measurable above the noise level, which for now is good enough for now. I do plan to make it better in the future I find these exchanges interesting and somewhat reassuring that I have considered, tested and eliminate at least the standard typical type of problems. Note all the issues that you have raised up to now are general and apply to standard types of configurations. Can you point out some specific issues that I need to be concerned with and may of missed that are unique to this configuration? As always, Thanks for your the comments and feedback ws *********************** > WarrenS wrote: >> ws reply to ">" Bruce's comments >> >> >>>> That calibration is linear over > than a 1 Hz (1e-7) offset range. >>>> >>> Whilst that may be true for your OCXO, this is certainly >>> not true for every ocxo one may wish to measure. >>> >> It is not the measured OSC that needs to be linear, >> but the reference Osc, but I'm sure you knew that. >> And NO not all reference Osc will be that good, >> but close enough, probable by an order of magnitude >> when proper end to end calibration is correctly done. >> If the range is small enough, it will be linear enough even >> for a Very nonlinear EFC curve. >> This is built to check High precision 10 MHz Osc, >> If they are off by more a Hz or even 1e-9, you don't need this kind of performance. >> > Not true, if the offset is accurately known then a stable oscillator > that has a frequency offset > 1E-9 is just as useful for calibration > purposes as one that has an offset of 1E-11 or less. >> This will easy give 1% absolute difference accuracy at 10 MHz. >> So the freq difference for a 1e-10 reading could be from 0.99e-10 to 1.01e-10. (plus noise) >> Most would consider it great to be able to take ADEV numbers that are not off more than two to one. >> >> >>> Since mixer nonlinearity near zero phase isn't an issue with a quiet ocxo >>> pair, a traditional diode double balanced mixer phase detector would be >>> appropriate as it has lower noise than any other mixer/phase detector. >>> >> With the high negative feedback that this has (2K+), the mixer hardly leaves 'zero' even with a noisy pair. >> I'm using a standard Mini-circuit DB Mixer. >> The mixer is working good enough so as not to be providing any noticeable error or noise. >> > Have you actually measured the mixer output noise when used as a phase > detector? >> I've tried two other ones and they all give about the same results. >> > One can either use them with both the RF and LO ports saturated and > suppress AM noise or only saturate the LO port. > The latter mode of operation tends to have a lower mixer phase shift > tempco with a lower phase sensitivity than when both the LO and RF ports > are saturated. > Phase detector IF port termination also affects its gain and noise. > Terminating the IF port in a capacitive load reduces the noise and > increases the phase detector gain. >> The Mixer is inside the Loop with a lot of negative feedback around it, >> Much of its basic characteristics and error are reduced by the loop gain, >> including much of its noise. The system noise is mostly determined by >> Just the EFC input noise and little else. >> The effective isolation between Oscillators with the mixer output RF shorted and the 10dB attenuators, >> is good enough not to cause any Osc interaction. >> >> > I doubt it. > Try estimating the required isolation using Adler's equation. >>> The isolation transformers are not useful for blocking RF, >>> >> Right, they pass RF, they don't block it. BUT >> The Isolation transformers are VERY important for a couple of reasons. >> The way these double oven 10811 are made, Their RF output is not isolated, >> but grounded at the inner case. At the end of the well shielded but long RF terminated cable >> there is a lot of RF noise on the cable shield ground AND a few mV of DC. >> If this Shield ground is Not isolated but connected to anything, It can cause a lot of errors. >> These errors can be orders of magnitude worse than other error sources. >> The RF shield ground on this OXCO can not be used in a dual Osc system, >> to get anywhere near max possible performance. It must be isolated. >> >> >>> nor in general is an arbitrary bypass capacitor array. >>> >> The circuit needs bypass caps because of the low level, wide bandwidth, high gain, "DC", feedback loop. >> Even when isolated there is enough RF on the ground shield that it still acts >> pretty acts like a 40 db down transmitting antenna. >> That RF needs to be kept out of the uV wide band signals. (And the other Osc) >> >> >> >>> I'd still prefer to compare results with and without the isolation amplifiers. >>> >> Yes, and I agree for a good safe universal GP design, They need to be there. >> I just don't happen to have any yet, and they are low on my priority list of needed improvements. >> With my specific setup, I have tested it well enough that I'm sure that their >> effect at this time would be small enough as not to be noticed. >> >> >>> If there is another contributor to phase locking then the full frequency >>> instability wont be reflected by the EFC input fluctuations. >>> >> Don't get me wrong, in the beginning the oscillators pretty much acted like they were sync locked >> because of poor ground, common PS, non isolated RF output, etc,etc. That did not go away by luck. >> What I'm saying is they do not now know that each other exist to the level that I can test to, which is about 5e-13. >> >> > >> Something that some seem to of missed is that this is not the standard RF circuit configuration with the standard open loop errors. >> This is a closed loop Neg feedback "PID type" freq control system where the errors inside the loop are reduced by the loop gain. >> The Only significant error outside the loop is the Osc its self and Osc's internal EFC offset. >> The EFC range being used during a measurement period is typically less than 1/1000 of its range. >> >> ws >> >> **************** >> ----- Original Message ----- >> From: "Bruce Griffiths" <bruce.griffiths@xtra.co.nz> >> To: "WarrenS" <warrensjmail-one@yahoo.com>; "Discussion of precise time and frequency measurement" <time-nuts@febo.com> >> Sent: Friday, October 02, 2009 6:00 PM >> Subject: Re: [time-nuts] Low cost alternate to Dual Mixer/DMTD >> >> >> >>> WarrenS wrote: >>> >>>> ws Reply to Bruce >>>> >>>> >>>>> You also need to measure the EFC slope at the operating point as the EFC >>>>> transfer characteristic can be highly nonlinear. >>>>> >>>>> >>>> Yes there is lots of things that can be done wrong but >>>> Another one of this configuration's many advantages is that the operating >>>> range of both the EFC and the Phase detector is very small, >>>> typical under a millivolt, so nonlinearly is NOT a problem. >>>> To calibrate end to end so that everything is included, The DUT can be changed by a >>>> small known offset, of say 1e-8 and measure the voltage change at the DVM/ADC output. >>>> Mine is calibrated for 1 mV per 1e-10 at the EFC, >>>> That calibration is linear over > than a 1 Hz (1e-7) offset range. >>>> >>>> >>> Whilst that may be true for your OCXO, this is certainly not true for >>> every ocxo one may wish to measure. >>> Since mixer nonlinearity near zero phase isnt an issue with a quiet ocxo >>> pair, a traditional diode double balanced mixer phase detector would be >>> appropriate as it hs lower noise than any other mixer/phase detector. >>> >>>>> Also need to ensure that injection locking doesn't occur >>>>> through injection via the EFC input. >>>>> >>>>> >>>> Yes, One of the reasons for the isolations transformers (and lots of bypass caps). >>>> >>>> >>> The isolation tranformers are not useful for blocking RF, nor in general >>> is an arbitrary bypass capacitor array. >>> >>>> If there is ANY ground noise between the Oscillators it can effect the EFC voltage. >>>> Like all low level uV signal measurement and control, a lot of attention HAS to be paid >>>> to insure there is no added noise or errors. This takes good analog and digital understanding >>>> of possible noise sources. Differential input and output amps go a long way to insure >>>> there is no ground loops, offset voltages or noise coupled errors. >>>> >>>> >>> These measures have little or no effect on RF injection via the EFC input. >>> Effective filtering of the phase detector output and use of a high >>> isolation mixer low noise is required. >>> >>>>>> Maybe due to the fact that the Osc are locked. >>>>>> >>>>>> >>>>> That is the worst possible case for injection locking. >>>>> >>>>> >>>> Possible, but not exactly what I've seen in this configuration. >>>> The effect of coupling between Oscillators is very phase sensitive, >>>> and can be positive, neg or null as their relative phase shifts. >>>> By adjusting the phase in the way I noted any changing effects >>>> can be easily seen when there is ANY interaction between Osc. >>>> >>>> >>>> >>> I'd still prefer to compare results with and without the isolation >>> amplifiers. >>> >>>>>> This is verified by adding a slow low level freq modulation on the DUT >>>>>> >>>>>> >>>>> Probably not a good test for injection locking as a small shift in >>>>> frequency from equality rapidly attenuates the effective injection >>>>> locking signal. >>>>> >>>>> >>>> Another one of the advantages of this circuit is that there is NO shift in freq >>>> from equaqlity (If that means what I think it is). In any case, >>>> I have found this to be a good test because ANY coupling of any type between >>>> Oscillators causes a nonlinear transfer function, as a function of voltage and/or freq. >>>> So by checking that the transfer function between the DUT EFC input >>>> and the fast ADC output is linear and freq independent over a wide range of signals, >>>> non coupling is assured down to the level of the noise. >>>> Another check I did was to unlock the two Osc and add an freq offset, >>>> to see if that caused ANY effect at all on the other Osc. >>>> >>>> >>>> >>> If there is another contributor to phase locking then the full frequency >>> instability wont be reflected by the EFC input fluctuations. >>> >>>>> Better test is to insert a very high reverse isolation amplifier between >>>>> each ocxo and the mixer and see if that makes any difference. >>>>> >>>>> >>>> Maybe so, and I did try to do that per your early suggestion, but the test was unsuccessful >>>> because of my poor isolation/buffer amps. It's something I'll get back to when >>>> I've lower the noise more to see if there is anything below the present noise level. >>>> >>>> ws >>>> >>>> *********************** >>>> From: "Bruce Griffiths" <bruce.griffiths@xtra.co.nz> >>>> >>>> >>>> >>>> >>> Bruce >>> >>> >>> >>> >> >> > >